A Step-by-Step Framework for Responsible AI Adoption in Education
AI could be the most useful thing to happen to Indian education in years, but only if we are honest about which problems we are actually trying to solve.
The problem with the rush to adopt AI
Every few years a new technology shows up promising to fix education. Right now it is artificial intelligence, and the pitch sounds familiar. Faster marking. Practice that adapts to each child. Less paperwork, so teachers finally get time to teach.
Spend a day in an average Indian school, though, and the picture is messier. Teachers are already juggling several apps, a steady flow of assessments, platforms that do not talk to each other, and reports due by Friday. And we are about to hand them one more tool.
So it is worth stopping to ask a few honest questions. Is this actually lightening the load, or quietly adding to it? Does every classroom need another digital intervention? Are schools bringing in AI because they have found a real learning problem to solve, or simply because AI is the thing everyone is talking about?
Underneath all of them sits one bigger question. Are we solving the right problem, or just adding more technology to a system that is already stretched? There is a real difference between putting what a school already does onto a screen and genuinely rethinking how it works. Digitising a broken process only gives you a faster broken process.
The OECD’s Education Policy Outlook 2024 makes a related point. AI can ease the administrative burden on schools, yet in many classrooms its promise for actual teaching and learning has still not shown up. The tool arrived. The improvement did not automatically follow.
OECD, Education Policy Outlook 2024. Click here to read the paper
The hidden cost: the thinking load on teachers and students
The workload problem is the visible one. There is a second cost that is easy to miss, and it is not about how busy a tool keeps people. It is about critical thinking and who ends up doing it. Start with teachers. AI does not just add another login, it changes the nature of the work.
The more a school leans on AI to produce, the more it quietly depends on teachers to catch what the machine gets wrong.
Students carry a sharper version of the same risk.
When AI hands over a clean answer too quickly, a child can skip every step of that effort and still turn in tidy homework. The work looks done. The learning never took place.
This is the line that matters, and it is the one to hold on to. AI should support the thinking that teachers and children do, never stand in for it.
Get that principle right and AI becomes genuinely useful. Get it wrong and you have automated away the very effort education exists to build.
To read the full paper, click the link:
UNESCO, Guidance for Generative AI in Education and Research (2023). Read the guidance
NCERT, National Curriculum Framework for School Education (2023). Access the framework
AI is far more than a teaching assistant
Here is where the conversation needs to widen. Most writing on AI in schools stops at lesson plans, worksheets and chatbots. Useful, but small. The bigger opportunity is to treat AI as a decision support layer, something that helps the people running education see more clearly and choose better, at every level of the system.
In the classroom, AI could help a teacher make sense of assessment results, spot the misconception that half the class shares, group students for different activities, and work out who genuinely needs extra time this week.
Across a student’s journey, learning analytics could follow individual progress over months rather than a single test, surface difficulties that keep recurring, and flag patterns early enough to act on.
Step up to the school, and the same approach helps a head teacher read attendance patterns, see where teachers are stretched thin, understand what the timetable will and will not allow, and notice the operational bottlenecks that repeat term after term.
Step up again, to the district and the state, and connected data could help administrators see which schools are struggling, anticipate where resources will be needed, and send academic support to the places that need it most, instead of spreading it evenly and hoping.
Notice what AI is doing in every one of these, and what it is not. It surfaces the pattern, sharpens the question, lays out the options. It does not decide. The teacher still chooses how to reach that child, the head teacher still fixes the timetable, the administrator still weighs where the support should go. AI is a tool, not the means to an end. The brain behind every one of these decisions is human, and it has to stay that way. A system that forgets this stops using AI to think better and starts letting it think instead.
This systems layer is where we spend most of our time at Sterlite EdIndia Foundation. Strengthening institutions, developing teachers, and making sense of learning data across the states where we work. It is unglamorous work, a long way from the shiny classroom app, and it is exactly where better information changes what actually happens to a child. The OECD’s 2025 work on AI adoption looks at this same territory, predicting enrolment, optimising timetables, allocating resources, while staying honest about what it means for the teachers at the centre of it all.
To read the full paper, click the link: Read here
Diagnose the problem before you deploy the technology
If there is one principle to hold on to, it is this. Start with the problem, not the product.
Picture a school where Grade 3 children are struggling to read. The reflex today is to reach for an AI reading app. But the app is the last question, not the first. Before anything else, the school needs to understand what is actually going on.
Which reading skills are weak? Is the difficulty concentrated in a few children, or spread across the whole class? Is something else feeding it, language, attendance, how reading is being taught, whether the children even have enough books to practise with? What could the teacher realistically do about it, starting Monday? And how will anyone know, a month from now, whether it worked?
Once those answers are on the table, AI can earn its place. It might help analyse the assessment patterns, suggest practice pitched at different levels, or track whether progress is actually coming. But if the real problem is too few teachers, too little instructional time, or children being taught in a language they do not yet follow, no application is going to fix that. It will only sit on top of the problem.
So before adding another platform, it is worth tidying the ones already there. Rationalise the existing tools, the logins, the reporting. A simpler system that people actually use beats a clever one that quietly gets abandoned.
Fix the foundations before scaling AI
India’s schools do not start from a blank page, and they do not share a single problem. The challenges are multi-layered and vary from one school to the next, so a single solution stretched across all of them rarely fits any of them well. A school short of teachers is facing a different question from one with crumbling infrastructure, and both differ again from a school where the building is fine and the staff are in place but learning outcomes are still slipping. The first needs people, the second needs investment in its premises, the third needs a sharper read on what is actually going wrong in the classroom. An all-encompassing package aimed at everything serves none of them, because it answers problems a given school may not have while missing the one it does. What each needs is a solution targeted at its specific gap.
That is why the order of things matters as much as the ambition. Before a school reaches for any tool, it helps to move through a few steps in sequence, each one earning the next.
How this plays out for one teacher
Take a teacher who notices, over a few weeks, that several children in her class keep getting subtraction wrong. This framework is how she moves from that worry to a decision she can trust.
First, she names the problem clearly. Rather than a vague sense that “the class is weak at maths,” she pins it down to subtraction with borrowing and sets a baseline from a short test, so she knows exactly where the children are starting and what progress would look like.
Next, she takes an honest look at what she already has. She goes back over the worksheets, the past assessments and the digital tools already in use, before reaching for anything new. Often the thing that helps is already on the shelf.
Then she asks what kind of problem this really is. Is it a teaching problem that needs a different explanation, a practice problem that needs more repetition, or something an AI tool could genuinely help with? Naming the type of problem is what points her to the right response.
Only then does she bring in AI, and only where it earns its place. She uses an AI tool to generate a differentiated subtraction worksheet, pitched at the levels her children actually need, which saves her an evening of preparation and lets her tailor the practice to the class.
Finally, she checks the results before scaling anything up. She compares the children’s performance against that first baseline, confirms the approach is actually working, and only then considers using it for other topics or other classes.
At no point does the tool make the decision. It sharpens the question and saves her time. The judgement, the teaching and the care stay with the teacher, which is exactly where they belong.
It is tempting to reach for examples from abroad, and international evidence can be genuinely useful. But a model that works in one country rests on its own infrastructure, languages, curriculum, teacher capacity and governance. None of that travels automatically. India’s approach has to be shaped by Indian classrooms, with teachers helping to choose, test and judge the tools they will be the ones to live with.
This section draws on an integrative literature review of teacher professional development for a future with generative AI. To read the full paper, click the link: Read here
This is roughly where the serious guidance lands too. UNESCO argues for approaches built around people, suited to the age of the child, and careful with privacy. The OECD makes the case for adoption that respects a teacher’s workload and leaves room for teachers themselves to experiment.
What this looks like in practice
Take a child who keeps stumbling on the same thing, say subtracting with borrowing, test after test. Instead of guessing, a teacher could look back over that child’s answers and see exactly where the method breaks down. From there she selects the right practice and tracks whether it actually closes the gap. The analysis does not teach the child. It shows the teacher where to direct her attention, which is the scarcest resource she has in a class of fifty.
Consider the end of every month, when teachers spend entire evenings consolidating data into reports for the administration. If the routine processing and summarising happened quietly in the background, those hours would go back where they belong, to planning lessons and sitting with the students who need it.
Then zoom out to a district deciding where to send its limited mentors, training and funds. By reading learning trends alongside the fuller picture of each school, its staffing, its attendance, its circumstances, administrators could stop spreading support thinly across everyone and start directing it to the schools that genuinely need it most.
None of this is something AI does on its own. Each case needs reliable data, a clear workflow, a person reviewing what comes out, and honest evidence that the intervention worked. The technology narrows the question. People still answer it.
Measure better decisions, not more technology
When people talk about digital transformation in schools, they tend to count the wrong things. How many AI tools were adopted. How many logins were created. How many dashboards now glow on a screen. None of that tells you whether a single child is learning any better.
In the end, the value of AI is not in how much of the system it can automate, but in whether it helps teachers and institutions make better decisions, while protecting the human relationships and the hard mental effort that learning depends on.
Working with Sterlite EdIndia Foundation
This is the work we care about. At Sterlite EdIndia Foundation, we work on strengthening systems, building capacity, and turning learning data into decisions that reach the children who need them most, across the states where we work. We also run courses on our akrava app that build a foundational and ethical understanding of AI, because getting this right means investing early in how AI is used in the classroom, not just whether it is. If you want to approach AI for learning rather than for dashboards, we would like to hear from you, and that is a conversation we are glad to have.
Sources and further reading
- UNESCO, Guidance for Generative AI in Education and Research (2023). Read the guidance
- OECD, Education Policy Outlook 2024. Read the report
- Artificial Intelligence in Teaching and Teacher Professional Development (2025), systematic review. Read the paper
- NCERT, National Curriculum Framework for School Education (2023). Access the framework
- OECD, AI Adoption in the Education System (2025). Read the report